Skip to main content

View Post [edit]

Poster: CR Saikley Date: Mar 22, 2004 3:29am
Forum: petabox Subject: Node Thermal Testing


Attached is the thermal data I collected on the prototype red boxes. Two units were stacked ontop of each other, insulated with styrofoam, fans were occluded, and openssl & iozone tests were run. A few observations:

Occluding the fans made a HUGE difference in internal temps. It's going to be critical to get airflow at the rack level.

Openssl with occluded fans caused dangerously high temps on the Via CPU. The temp spec isn't in Via's literature, but at 195F (!) I'm sure that we exceeded it.

The disk temps didn't budge when the CPU got really hot. The airflow (and relatively low thermal mass) is preventing the CPU from heating the disks.

Iozone induced roughly a 7 deg C rise in disk temp.

The single exhaust fan near the CPU doesn't help. Temperatures are approximately the same with or without it. It also doesn't matter whether that fan is exhausting out the front or the side.

Presence or absence of the NIC does not have a significant effect on disk or CPU tempertaure.

The box on top was definitely hotter than the box below. This is partly due to the lower box heating the upper one, and partly due to the fact that the upper box had no airflow beneath the disks.

Airflow under the disks is significanly more effective that airflow above the disks - about 5 degree C difference.

The disk temperature rose by 3-5 degrees C just from being stacked on top of another machine. This will be something to watch very carefully when we have a full rack of this stuff.


Attachment: Temp Tests1.xls

Reply [edit]

Poster: CR Saikley Date: Mar 29, 2004 3:08am
Forum: petabox Subject: Re: Node Thermal Testing

Additional thermal tests were performed on 4 stacked systems. The following was observed:

The temperature rise when stacking 4 units was equal to the rise when stacking two. This is was the expected results, and bodes well for stacking 40 machines.

The disk temperatures in units 3 & 4 stayed much cooler than those in units 1 & 2. We do not yet know why.

Attachment: Temp Tests1.xls

Reply [edit]

Poster: CR Saikley Date: Mar 23, 2004 9:17am
Forum: petabox Subject: Re: Node Thermal Testing

Additional measurements were made today of input versus output air temperatures. Temperature deltas were 10-12 degrees F. Ten degrees deltas were observed on idle machines (disks spinning) and 12 degree deltas were observed on openssl test. Iozone tests yielded an 11 degree rise.

Data are attached.

Attachment: Temp Tests1.xls